Language Edition

Two speakers share a short scene. One edition replaces a line in its original language; another gives the scene translated dialogue and captions. Both must return to the same locked source. Someone must answer for the words, casting, performance, timing, disclosure, and accepted result. Language Edition keeps that work together.

A language edition need not translate. Same-language replacement, narration, repair, and accessibility belong to the same durable project. Its passage ends in an attributable set of timed-language assets, with a constrained track replacement when requested.

Contract

The Composition is language_edition.timed_media revision 2; its principal Pattern is language_edition.build_language_version@2. A LanguageVersionRequest@2 states:

• Source: an immutable media reference and timebase, or an exact locator with source-eligibility criteria and requested custody posture.
• Meaning: source and target languages, localization brief, glossary, and intended audience.
• Performance: exact eligible performer, voice, and capability references, with rights and disclosure requirements.
• Finish: timing, captions, format, accessibility, and acceptance conditions.

The request gives every later revision a common task. Generic translation, speech engines, network acquisition, byte custody, Persona identity, music, visuals, editorial recutting, and publication retain their own owners. Crossing into their work requires separate admission.

One bounded language-edition road

Establish the source. Language Edition judges whether the source fits the request. A locator may require a separately granted Scout Fetch or Download. Scout owns Search, Fetch, Download, destination and contact policy, quarantine arrival, and acquisition receipts. After bounded quarantine, separately granted, custody-owned Artifact Admission validates retained bytes and establishes the immutable source master. A URL cannot stand in for that master; an acquisition receipt cannot settle custody.

Establish what was spoken. Bounded demux and probe precede Echo Ear's transcript, diarization, and timing observations. The proposed DialogueTranscriptRevision@2 remains observation; an authorized source-language reviewer must judge it. That reviewer creates SourceDialogueAttestation@1, binding the exact source and timebase, stable segment identities, speakers, words, timing, uncertainty and corrections, and the reviewer's Principal and evidence. Ambiguity stays unresolved. Only then may source-aligned wording become canonical. When Broadcast's ScriptRevision@1 supplies canonical words, the attestation binds them to spoken segments without revising or taking ownership of that script.

Prepare and review the edition. The attested source enters an optional attributed Translation Spell and mandatory review, producing LanguageEditionScriptRevision@2. A same-language edition can proceed without translation. Language Edition judges the wording in its scene; the translation contract keeps the transformation inspectable.

Perform and conform. Eligible casting selects human takes or Echo Voice synthesis for DialoguePerformanceRevision@2. Preserve performer and voice eligibility: selecting them creates no grant and copies no Persona authority. Dialogue editing, captions, and DialogueConformMap@2 align the performance to the source. Review linguistic fit, intelligibility, timing, rights, and disclosure before acceptance.

Hand over the assets. TimedLanguageAssetBundle@2 binds exact source, attestation, and script revisions; performance and dialogue artifacts; captions and conform map; rights, disclosure, findings, lineage, and admission criteria. A consumer can admit this bundle on its own. It is neither a final picture container nor an editorial recut.

End where the evidence ends. If requested, restricted track mux and probes with back-check may produce LanguageVersionCandidate@2 under the locked-picture contract. LanguageEditionFindingSet@2 carries the findings. Each edition retains its attested source, its own language and performance revisions, and every artifact actually produced. Partial, incomplete, and unknown outcomes remain exact; an optional mux does not become canonical merely because it exists.

Translation is a Spell, localization is the judgment

The authority-qualified publisher of a versioned Translation Spell owns its transformation semantics; Spellweaver validates its placement. The request pins attested text, languages, glossary and protected spans, implementation, detection and fallback, review, and declared loss. Retain original and derivative so the edition's reviewer can judge speaker, scene, audience, duration, and fit.

Language detection grants no translation permission. Provider failure grants no substitute language, voice, model, execution road, or paid service. Another Composition can reuse the exact Spell while retaining its own judgment.

Language Edition owns captions, subtitles, scripts, and on-screen-language cue sheets as parts of a timed edition. Content owners retain untimed writing; Frontend retains web and Altar chrome catalogues. English-only Altar reference chrome requires neither English media nor translated canonical identities.

Voices belong to no shortcut

Echo owns capture, STT, TTS, voice-profile facts, delivery, and playback chronology. The Animator profile identifies the exact engine, model and acoustic voice, supported languages and formats, licence, and measured limits. Performer consent and source rights must independently cover the exact inputs.

Language Edition casts and directs pronunciation and performance, then accepts timing and disclosure. Avatar may supply exact eligibility for a Lich voice while retaining Persona-linked presentation and Morphe. Resemblance establishes neither identity nor rights; approval of one line does not authorize arbitrary later words. Riffmaw retains sung performance and music. Across every road, generated sound proves neither playback nor hearing.

Preserve the locked picture

The optional mux may add or replace only declared dialogue, narration, caption, subtitle, or language-metadata tracks. Preserve every non-language stream. Record the preserved and replaced digests, timebases, codecs, channels, and offsets, the loudness and intelligibility probes, and the result digest. These records let a reviewer check what survived and exactly what changed.

Changed cuts, scene order, claims, picture sound, or music return to Broadcast. A constrained language replacement cannot quietly become a new edit. The asset bundle remains a valid handoff without the optional mux; each receiving owner judges the exact result for its own use.

Model-native timed-language facet

A VideoJob@2 may emit synchronized language only when its pre-execution MediaFacetAuthoritySet@1 declares timed_language, language_edition.timed_media revision 2, and the exact LanguageVersionRequest@2 digest. Before execution, resolve the exact Echo speech interface and profile, performer and acoustic voice, language support, consent, rights, disclosure, and retention. Native video audio has the same obligations.

After Prism settlement, Language Edition independently issues SemanticFacetAdmissionReceipt@1 over child and compound digests, source attestation and script, speaker and language role, use, rights, timing, findings, decision maker, and disposition. The bundle and optional mux retain that receipt and parent. Undeclared speech or text is quarantined or deleted. Demux, labels, or another owner's receipt cannot admit the facet. An inseparable compound needs every required owner's acceptance of the exact digest and use.

Neighbours keep their truth

The source and the result pass through distinct kinds of judgment:

Settled exact references need no Suite. A promised result that actively admits, awaits, cancels, retries, recovers, or jointly settles new work across those owners requires one. A Product may package a language edition without taking its records; the technical Composition name does not thereby become the market name.

Failure, restart, and the first proof

Stable segment identities bind the exact source and timebase. Transcript, attestation, translation, performance, alignment, dialogue-mix, mux, and review attempts keep independent revisions and dispositions. Unsupported languages, ambiguous speakers, missing rights, timing loss, failed intelligibility, stale source, budget exhaustion, and indeterminate provider or write effects produce the appropriate findings, refusal, partial result, or unknown outcome. Restart uses reconciled immutable artifacts and settled attempts, without silent substitution.

The first falsifying proof remains a target: run a network-disabled synthetic two-speaker video with separate picture, dialogue, music-effects, and caption streams on a known timebase. Fake Ear, Translation, and Voice should build both a same-language and a translated edition. Check preserved picture and non-language digests, lineage, attestation after human correction, glossary protection, timing pressure, caption alignment, partial speaker refusal, crash recovery, deterministic mux, export, and deletion. This would prove the contract's passage. It would not prove translation quality, real-voice performance, rights, publication, or delivered operation.

Workflow owns the score and its admissions; Audio and Vision own the media boundaries. Return to the Composition Portfolio for the applications that may receive the edition.